Interesting–I had just been looking at a model (graphviz below) to get group means for psi, beta, and alpha (again, I’m fairly new at this, so not sure that what I’m doing is exactly correct but it seems to be working upon first glance). My plan was to add the group-level random effects in, just hadn’t gotten there yet. I was able to increase the data while simultaneously decreasing the sampling time, and there were 0 divergences. The only hiccup was that since I completely decoupled the Bernoulli from the Gamma, if a group doesn’t have any “success” events, I don’t get any estimate on the Gamma portion (noted by groups on the left being 287 and the right being 250). Potentially there is a way for me to leverage your CustomDist into what I’ve done–I think with everything I know now I’ve at least got a couple of things to try. Also open to feedback if anything in my graphviz looks whacky!
